Vehicle-based Thermal Imaging Target Detection Method Based on Enhanced Lightweight Network
-
摘要: 车载热成像系统不依赖光源,对天气状况不敏感,探测距离远,对夜间行车有很大辅助作用,热成像自动目标检测对夜间智能驾驶具有重要意义。车载热成像系统所采集的红外图像相比可见光图像具有分辨率低,远距离小目标细节模糊的特点,且热成像目标检测方法需考虑车辆移动速度所要求的算法实时性以及车载嵌入式平台的计算能力。针对以上问题,本文提出了一种针对热成像系统的增强型轻量级红外目标检测网络(Infrared YOLO,I-YOLO),该网络采用(Tiny you only look once,Tiny-YOLO V3)的基础结构,根据红外图像特点,提取浅层卷积层特征,提高红外小目标检测能力,使用单通道卷积核,降低运算量,检测部分使用基于CenterNet结构的检测方式以降低误检测率,提高检测速度。经实际测试,Enhanced Tiny-YOLO目标检测网络在热成像目标检测方面,平均检测率可达91%,检测平均速度达到81Fps,训练模型权重96MB,适宜于车载嵌入式系统上部署。
-
关键词:
- 车载热成像系统 /
- 夜间智能驾驶 /
- I-YOLO红外目标检测网络 /
- CenterNet结构 /
- 车载嵌入式平台
Abstract: A vehicle-based thermal imaging system does not depend on a light source, is insensitive to weather, and has a long detection distance. Automatic target detection using vehicle-based thermal imaging is of great significance for intelligent night driving. Compared with visible images, the infrared images acquired by a vehicle-based thermal imaging system based on existing algorithms have low resolution, and the details of small long-range targets are blurred. Moreover, the real-time algorithm performance required to address the vehicle speed and computing ability of the vehicle-embedded platform should be considered in the vehicle-based thermal imaging target detection method. To solve these problems, an enhanced lightweight infrared target detection network (I-YOLO) for a vehicle-based thermal imaging system is proposed in this study. The network uses a tiny you only look once(Tiny-YOLOV3) infrastructure to extract shallow convolution-layer features according to the characteristics of infrared images to improve the detection of small infrared targets. A single-channel convolutional core was used to reduce the amount of computation. A detection method based on a CenterNet structure is used to reduce the false detection rate and improve the detection speed. The actual test shows that the average detection rate of the I-YOLO target detection network in vehicle-based thermal imaging target detection reached 91%, while the average detection speed was81 fps, and the weight of the training model was96MB, which is suitable for deployment on a vehicle-based embedded system. -
表 1 4类检测目标统计数据分析
Table 1. Statistical analysis of four kinds of detection targets
Detection model Mp(%) Mf(%) Mm(%) Person Car Bus Truck Person Car Bus Truck Person Car Bus Truck SSD300×300 66 71 73 68 12 13 14 11 21 12 21 20 RetinaNet-50-500 90 89 88 92 15 17 18 14 6 4 6 14 Tiny-YOLOV3 65 70 75 69 15 10 15 10 20 15 23 21 YOLOV3 95 90 90 95 20 18 20 15 5 3 5 15 I-YOLO 91 88 89 93 3 5 3 5 9 8 6 18 表 2 综合性能测试对比分析
Table 2. Comparison and analysis of comprehensive performance tests
Detection model Mp/% Mf/% Mm/% Mo/FPS Mw/MB SSD300×300 67 11 31 13 196 RetinaNet-50-500 90 15 13 7 246 Tiny-YOLOV3 66 12 32 62 34 YOLOV3 95 16 6 21 234 I-YOLO 91 6 12 81 96 -
[1] 崔美玉. 论红外热像仪的应用领域及技术特点[J]. 中国安防, 2014(12): 90-93. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGAF201412044.htmCUI Meiyu. On the Application Field and Technical Characteristics of Infrared Thermal Imager[J]. China Security, 2014(12): 90-93. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGAF201412044.htm [2] 范延军. 基于机器视觉的先进辅助驾驶系统关键技术研究[D]. 南京: 东南大学, 2016.FAN Yanjun. Research on Key Technologies of Advanced Auxiliary Driving System Based on Machine Vision[D]. Nanjing: Southeast University, 2016. [3] 杨阳, 杨静宇. 基于显著性分割的红外行人检测[J]. 南京理工大学学报: 自然科学版, 2013, 37(2): 251-256. doi: 10.3969/j.issn.1005-9830.2013.02.009YANG Yang, YANG Jingyu. Infrared Pedestrian Detection Based on Significance Segmentation[J]. Journal of Nanjing University of Technology: Natural Science Edition, 2013, 37(2): 251-256. doi: 10.3969/j.issn.1005-9830.2013.02.009 [4] LE Cun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278- 2324. doi: 10.1109/5.726791 [5] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//IEEE Conference onComputer Vision and Pattern Recognition, 2014: 580-587. [6] HE K M, ZHANG X Y, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916. doi: 10.1109/TPAMI.2015.2389824 [7] Girshick R. Fast R-CNN[C]//IEEEInternational Conference on Computer Vision, 2015: 1440-1448. [8] REN S Q, HE K M, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. doi: 10.1109/TPAMI.2016.2577031 [9] LI Y, HE K, SUN J. R-FCN: Object detection via region-based fully convolutional networks[C]//Advances in Neural Information Processing Systems, 2016: 379-387. [10] LIU W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector[C]//European Conference on Computer Vision, 2016: 21-37. [11] Redmon J, Farhadi A. YOLO9000: Better, faster, stronger[C]//IEEE Conference on Computer Vision and Pattern Recognition, 2017: 6517- 6525. [12] Redmon J, Farhadi A. Yolov3: An incremental improvement[EB/OL]. (2018-04-08)[2018-09-07]. https://arxiv.org/abs/1804.02767 [13] ZHANG Y, SHEN Y L, ZHANG J. An improved Tiny-YOLOv3 pedestrian detection algorithm[J]. Optik, 2019(183): 17–23. http://ieeexplore.ieee.org/document/8868839 [14] DUAN Kaiwen, BAI Song, XIE Lingxi, et al. CenterNet: Keypoint triplets for object detection[C]//Proceedings of the 2019 IEEE International Conference on Computer Vision. NJ: IEEE, 2019: 6569-6578. [15] 吴天舒, 张志佳, 刘云鹏, 等. 基于改进SSD的轻量化小目标检测算法[J]. 红外与激光工程, 2018, 47(7): 703005-0703005(7). https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ201807007.htmWU Tianshu, ZHANG Zhijia, LIU Yunpeng, et al. A lightweight small object detection algorithm based on improved SSD[J]. Infrared and Laser Engineering, 2018, 47(7): 703005-0703011. https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ201807007.htm [16] 唐聪, 凌永顺, 郑科栋, 等. 基于深度学习的多视窗SSD目标检测方法[J]. 红外与激光工程, 2018, 47(1): 126003-126011. https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ201801042.htmTANG Cong, LING Yongshun, ZHENG Kedong, et al. Object detection method of multi-view SSD based on deep learning[J]. Infrared and Laser Engineering, 2018, 47(1): 126003-0126011. https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ201801042.htm [17] 张祥越, 丁庆海, 罗海波, 等. 基于改进LCM的红外小目标检测算法[J]. 红外与激光工程, 2017, 46(7): 726002-0726008. https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ201707040.htmZHANG Xiangyue, DING Qinghai, LUO Haibo, et al. Infrared dim target detection algorithm based on improved LCM[J]. Infrared and Laser Engineering, 2017, 46(7): 726002-0726008. https://www.cnki.com.cn/Article/CJFDTOTAL-HWYJ201707040.htm [18] 张小荣, 胡炳梁, 潘志斌, 等. 基于张量表示的高光谱图像目标检测算法[J]. 光学精密工程, 2019, 27(2): 488-498. https://www.cnki.com.cn/Article/CJFDTOTAL-GXJM201902025.htmZHANG Xiaorong, HU Bingliang, PAN Zhibin, et al. Tensor Representation Based Target Detection for Hyperspectral Imagery[J]. Editorial Office of Optics and Precision Engineering, 2019, 27(2): 488-498. https://www.cnki.com.cn/Article/CJFDTOTAL-GXJM201902025.htm [19] 王洪庆, 许廷发, 孙兴龙, 等. 目标运动轨迹匹配式的红外-可见光视频自动配准[J]. 光学精密工程, 2018, 26(6): 1533-1541. https://www.cnki.com.cn/Article/CJFDTOTAL-GXJM201806028.htmWANG Hongqing, XU Tingfa, SUN Xinglong, et al. Infrared-visible video registration with matching motion trajectories of targets[J]. Editorial Office of Optics and Precision Engineering, 2018, 26(6): 1533-1541. https://www.cnki.com.cn/Article/CJFDTOTAL-GXJM201806028.htm